1 Select project(s)

On top left, click the dropdown widget. You may either a) scroll down, or b) type in keywords, to find your project(s) of interest.


Click and select your project(s) of interest. Then, click Confirm selection! to proceed.


To re-select project(s), click Reset selection.


2 Censoring

Censoring allows analysis with a specified time interval. You may choose a censoring time in days, months (30.4375 days), or years (365.25 days). For a selected clinical endpoint (e.g. overall survival (OS), progression-free survival (PFS)), if the time-to-event is larger than the defined time, we set censoring status to 0 and time to the defined time; if the time-to-event is smaller than or equal to the defined time, we set censoring status to 1 and time stays unchanged. By default, cases are censored at 10 years.


3 Single or interaction

Under No. of analysis:, enter or click the upper arrow button to 2 to analyze interactions between biomarkers. Choose 1 to analyze a single biomarker.


4 TCGA: clinical endpoint

If TCGA, select the clinical endpoint of interest. Recommendations are provided according to TCGA pan-cancer clinical study (Liu et al., Cell, 2018). You may also click on the lightblue book button for the full published recommendation table.


5 DepMap: cell lines & gene/drug

Using DepMap data, cSurvival can assess how a biomarker or combinations of biomarkers affect(s) the knockout (CRISPR) / knockdown (RNAi) effect of a gene, or the sensitivity to a drug compound. First, select the type of DepMap data to analyze:


Next, select the cancer type(s) to study. You may type in keywords to search your cancer type(s) of interest.


Optionally, filter the cell lines by subtypes. By default, all subtypes of the selected cancer type(s) are selected.


Optionally, manually select specific cell lines. By default, all cell lines of the selected subtype(s) are selected.


Then, type to search, and select your gene (or drug) of interest.


6 Select biomarker(s)

cSurvival supports survival analysis with a wide range of biomarkers including gene or GS expression, somatic mutation, microRNA expression, DNA methylation, and protein expression. In addition, two biomarkers can be selected at a time for interaction analysis.

Mandatory fields are marked with an orange asterisk.

6.1 Single biomarker

Make sure No. of analysis is 1. You will see a light grey panel in the center. First, select data category of interest.

6.1.1 Gene-level biomarker

Select Gene or locus to study if the expression level, mutational status, copy number, methylation, or protein level of a gene or locus correlates with poorer/better survival:


Next, select the type of molecular data to analyze:

  • Expression: a gene’s expression level
  • Mutation: a gene’s mutational status
  • CNV: copy number variation of a DNA segment
  • miRNA: a microRNA’s expression
  • Methylation: methylation level of a DNA segment
  • Protein: expression level of a protein


Then, type to search your gene of interest:


Make sure you have clicked on the gene so it is properly selected:


If “Expression”, option is provided to normalize the selected gene’s expression against another gene, or a gene set.

6.1.2 GS-level biomarker

Select Gene set to study if the average expression level of a gene set correlates with cancer survival, e.g. genes in the same pathway, TF targets, drug targets, miRNA targets, interacting proteins, or user-defined list of genes:


Next, select Library (default) to choose a gene set from eVITTA to analyze a pathway, a biological process, a cellular location, a transcriptional factor, a drug, or a gene’s interacting partners.


Then, select a database. For full list of available databases, visit easyGSEA User Guide:


After that, gene sets from the selected database will be loaded. Search for keywords (e.g. glycolysis, chemokine, signaling) and select the one of interest:


(Optional) You could filter out gene sets that contains your gene(s) of interest, in HUGO symbol format, delimited by “&” (and) or “|” (or). “|” is evaluated before “&”. Example: MYC&TP53|BCL2&BRCA1|BRCA2 is evaluated as MYC&(TP53|BCL2)&(BRCA1|BRCA2), which means MYC and (TP53 or BCL2) and (BRCA1 or BRCA2). A maximum of 10 genes are supported. Click the magnifier button to search and filter; click the cross button to cancel the filter.


(Optional) After filtering, only gene sets containing the gene(s) will be displayed:


Click to select your gene set of interest. Upon successful selection, feedback is provided showing the genes contained in the gene set. To the right, you could click the red download button to download the gene list, or click the red link button for more details about the selected gene set from its original database.




Alternatively, select Manual to enter your own list of genes.


Enter your genes in HUGO symbol format, delimited by newline, space or comma. Click Submit to submit your list. Feedback is provided upon successful submission.




6.2 Two biomarkers

Make sure No. of analysis is increased to 2. You will see two panels:


Select each biomarker in each panel. Below are guidelines for different combinations of biomarkers. For more detailed explanation for individual biomarker selection, see above 6.1 Single biomarker.

6.2.1 Gene and gene

Select Gene or locus in both panels:
For more detailed explanation for individual gene/locus selection, see above 6.1.1 Gene-level biomarker.

6.2.2 GS and GS

Select Gene set (GS) in both panels:
For more detailed explanation for individual GS selection, see above 6.1.2 GS-level biomarker.

6.2.3 Gene and GS

Select Gene or locus in one panel, and Gene set (GS) in the other:
For more detailed explanation for individual gene/locus and GS selection, see above 6.1.1 Gene-level biomarker and 6.1.2 GS-level biomarker.


7 Run parameters

Click the light-blue round gear button on top right of Confirm and analyze! for cSurvival’s refined and user-controlled parameters:

7.1 Common parameter

For TCGA, we removed 507 problematic cases based on comments in the merged_sample_quality_annotations.tsv file by default, but a user can choose to include them via the web interface:

7.2 Continuous variable

Parameters specifically for analyzing continuous variables (e.g. gene/GS/miRNA/protein expression, DNA methylation, cell line copy number) include:

  • Method to determine the minimum P-value (only applicable to dynamic iteration approach):


  • Number of permutations to calculate the false discovery rate (FDR) (default:100); increased permutations improve statistical precision, but longer run time may be expected:


  • We provide two alternative methods to stratify patients into risk groups: i) dynamic iteration with the minimum P-value approach, and ii) user-selected quantile. For i):


    For ii):

7.3 Categorical variable

For mutations, by default, variants to be classified as High/Moderate variant consequences are selected as the “Mutated” group, while wild-type (WT) and variants to be classified as Low/No variant consequences are selected as the “Other” group. Adjust to suit the purpose of your study. For more information about variant classification: http://uswest.ensembl.org/Help/Glossary?id=535:




For thresholded copy number data, select i) Automatic to automatically determines whether copy number gain and/or loss results in more significant survival difference, ii) Copy number gain to compare cases with copy number gain with the rest of the population, and iii) Copy number loss to compare cases with copy number loss with the rest of the population:




7.4 Joint analysis

In joint analysis with two genomic predictors, select the risk group (subgroup) of interest. If All is selected, an ANOVA-like test is done to test if there is significant difference between any of the four subgroups. If a specific subgroup is selected, it is compared against the other subgroups as a whole:


Also adjustable is the minimum number of cases (default, 10% of the population) to define a subgroup. This parameter only applies to dynamic iteration with continuous variables (e.g. mRNA gene expression, DNA methylation):


Another parameter for joint analysis with two continuous genomic predictors is the dynamic search method: Median-anchored greedy search (heuristic) determines the minimum P-value by finding the percentile in variable 2 that gives the minimum P-value on the median percentile in variable 1, then looking for percentile combinations that give lower P-values via greedy search. Exhaustive search determines the minimum P-value by testing all percentile combinations:



8 Visualization customization

8.1 Survival curves

We offer options to customize the survival curves:

8.2 Scatter plot

Customization options for a scatter plot:

8.3 Violin plot

Customization options for a violin plot:

8.4 Heatmap tracking

Customization options for a heatmap tracking looping percentiles in joint analysis with two continuous genomic predictors:


9 Application cases

9.1 Single biomarker analysis

9.1.1 Expression

To test if the expression level of RELA (v-rel avian reticuloendotheliosis viral oncogene homolog A) is correlated with overall survival in lung adenocarcinoma (TCGA-LUAD):


KM log-rank test is performed and its result is displayed by default:



Percentile tracking of the KM log-rank result:


To run the Cox regression test:


The Cox regression result:


Percentile tracking of the Cox regression result:

9.1.2 Mutation

To test if missense (potentially gain-of-function) mutation of EGFR (epidermal growth factor receptor) is correlated with overall survival in lung adenocarcinoma (TCGA-LUAD):


Adjust advanced run parameters to analyze missense mutations:


Missense mutation of EGFR is correlated with poor prognosis in TCGA-LUAD patients:

9.1.3 microRNA Expression

To test if the expression level of miR-130a-3p is correlated with overall survival in stomach adenocarcinoma (TCGA-STAD):


KM log-rank test is performed and its result is displayed by default:



Percentile tracking of the KM log-rank result:


To run the Cox regression test:


The Cox regression result:


Percentile tracking of the Cox regression result:

9.1.4 DNA methylation

To test if methylation level of APC (APC Regulator Of WNT Signaling Pathway) at the locus chr5:112737735-112737737 (cg16970232) is associated with cancer progression in prostate adenocarcinoma (TCGA-PRAD):




KM log-rank test is performed and its result is displayed by default:



Percentile tracking of the KM log-rank result:


To run the Cox regression test:


The Cox regression result:


Percentile tracking of the Cox regression result:

9.1.5 Protein expression

To test if the protein expression level of collagen VI is associated with overall survival in skin cutaneous melanoma (TCGA-SKCM):




KM log-rank test is performed and its result is displayed by default:



Percentile tracking of the KM log-rank result:


To run the Cox regression test:


The Cox regression result:


Percentile tracking of the Cox regression result:

9.1.6 GS expression

To test if expression of the nuclear factor-erythroid factor 2-related factor 2 (Nrf2)-antioxidant response element (ARE) pathway genes is associated with overall survival in lung adenocarcinoma (TCGA-LUAD):



KM log-rank test is performed and its result is displayed by default:



Percentile tracking of the KM log-rank result:


To run the Cox regression test:


The Cox regression result:



Percentile tracking of the Cox regression result:


9.2 Biomarker interaction analysis

9.2.1 Expression & expression

To test if the expression levels of solute carrier family 7 member 11 (SLC7A11) and solute carrier family 2 member 1 (SLC2A1, also known as glucose transporter 1 (GLUT1)) are jointly associated with overall survival in liver hepatocellular carcinoma (TCGA-LIHC):



By default, differences between all risk subgroups are tested by KM log-rank:



Percentile tracking of the KM log-rank result; optimal cutoff combination is highlighted by a yellow arrow:



Click Expression-Expression scatter to measure the correlation between expressions of SLC7A11 and SLC2A1:


To run the Cox regression test:


In joint analysis, detailed Cox regression models are provided to assess how two predictors jointly impact outcomes by calculating the effect sizes (hazard ratios, HRs) and the statistical significances of the two predictors and their interaction. We could see that SLC7A11 (P = 0.02040) and SLC2A1 (P = 0.00551) each had an effect on prognosis, while their interaction term was not significant (P = 0.89597), suggesting that they might function in separate pathways to affect prognosis:



Percentile tracking of the Cox regression result; optimal cutoff combination is highlighted by a yellow arrow:



From the ANOVA-like results above, we could tell that the High-High risk subgroup (patients with high expressions of both SLC7A11 and SLC2A1) showed a worse outcome than the other three subgroups (patients with low expressions of SLC7A11 and/or SLC2A1). Therefore, we selected the High-High risk subgroup to contrast against the combined other three. Adjust the advanced run parameters:


The High-High group showed a worse prognosis as assessed with KM log-rank:



Percentile tracking of the KM log-rank result; optimal cutoff combination is highlighted by a yellow arrow:




Such correlation is also true with Cox regression:



Percentile tracking of the Cox regression result; optimal cutoff combination is highlighted by a yellow arrow:




9.3 DepMap

9.3.1 CRISPR

To test in lung cancer cell lines, if the knockout effect by NFE2 Like BZIP Transcription Factor 2 (NFE2L2) is correlated with mutations of Kelch-like ECH-associated protein 1 (KEAP1):


KEAP1-mutated lung cancer cell lines were more sensitive to NFE2L2 knockout:

9.3.2 RNAi

To test in lung cancer cell lines, if the knockdown effect by NFE2 Like BZIP Transcription Factor 2 (NFE2L2) is correlated with mutations of Kelch-like ECH-associated protein 1 (KEAP1):


KEAP1-mutated lung cancer cell lines were more sensitive to NFE2L2 knockdown:

9.3.3 Drug sensitivity

To test in lung cancer cell lines, if the response to a potent oxidative stress inducer, menadione (BRD-K78126613-001-28-5), is correlated with the expression of genes in the nuclear factor-erythroid factor 2-related factor 2 (Nrf2)-antioxidant response element (ARE) pathway:


KEAP1-mutated lung cancer cell lines were more sensitive to NFE2L2 knockdown:


10 Support


Feel free to reach us at or if you have any questions!


Q&A - known issues

Q: I can’t click on any button or select any options. Seems the website is not responding. Why is that?
A: Two possible reasons. First, check whether your internet is stable. Second, we might be experiencing high traffic. Please come back after a few minutes. If problem still exists. Please email us.